Consider declaring max message size up front to get smaller offsets

By declaring the message to be less than 2^8, 2^16, or 2^32 bytes, all offsets and lengths could be stored as u8, u16, or u32. This could be a substantial savings.

Could do this via a crate feature.

Start by locating the crate's offset and length representation and its feature configuration. Determine how a declared maximum message size could select u8, u16, or u32 storage, then verify that serialization remains correct for each supported size tier.

Written by the indexing model from the issue text.